Evidence for transmission or disease reductions following ACT deployment initially came from studies in South East Asia [9–12] and Southern Africa [13] which recorded a reduction in population P. falciparum prevalence or disease incidence after ACT became the main treatment for clinical malaria in the area. Since then, data from a number of settings including Zanzibar [14] and Rwanda have demonstrated a significant reduction in malaria cases following ACT introduction. While these studies are consistent in suggesting a benefit of ACT, they had observational, time-trend designs without control groups, and it is unclear how much of the effect in these settings was attributable to ACT and how much to other factors including simultaneous introduction of vector control measures [11,13], changes in climatic conditions [10], and change in diagnostic quality control. Furthermore, these observations come from relatively low transmission settings. In higher-transmission settings, the asymptomatic reservoir of infection may limit the potential for transmission reduction by treatment of symptomatic cases, because greater exposure increases immunity and the chance of an infection remaining asymptomatic [15,16].

A previous mathematical model of symptomatic case management predicted moderate to substantial impact of the nonartemisinin-based antimalarials sulfadoxine-pyrimethamine (SP) and amodiaquine (AQ) on age-prevalence patterns of infection and rates of severe disease and mortality under scenarios of 40% or 100% coverage [17]. Rates of uncomplicated clinical malaria episodes were reduced in low but not high transmission settings. Understanding to what extent ACTs with their additional gametocytocidal effects can reduce transmission in settings where partially effective treatments are already in place would aid malaria control agencies who currently must choose how to divide resources between drug purchases, drug delivery strategies, and other control interventions. The decision to maximise the coverage of the ACT class of drugs beyond the formal health care sector has substantial cost implications given their higher price [18], but it would be supported if ACT were likely to have a substantial effect on transmission. Choosing to invest in improved diagnostic techniques can save on overtreatment of fever cases with ACT; however, the importance of this widespread presumptive treatment for keeping transmission levels down needs to be investigated.

With increasing investment in research, the range of drugs in development for treatment of malaria is widening and a greater number are also becoming available in combination with artemisinins as ACT [19]. The stronger gametocytocidal activity of artemisinin derivatives compared to previous first-line antimalarials [20] has received the most attention in relation to potential transmission reductions by treatment [7]. However, two other main drug properties impact on transmission: prophylaxis and efficacy at clearing parasitaemia. Artemisinin derivatives have a very short prophylactic time in relation to other antimalarials, and although they remain highly efficacious in most areas of the world due to lack of parasite resistance, there are nonartemisinin antimalarial regimens that are of comparable efficacy in some regions [21]. Knowledge of the relative importance of these different drug properties in reducing transmission could help guide policy makers in choosing between ACTs with different partner drugs, and in the longer term a replacement first-line treatment, in the eventuality of parasite resistance or prohibitive cost of ACTs.

Our analysis finds that gametocytocidal properties of antimalarials have the most relative impact in lower-transmission settings, while prophylactic effects have more impact in areas with high transmission. An ACT with the gametocytocidal effects of the artemisinin derivative and the prophylactic effect of a longer-acting partner drug is predicted to have the greatest impact on transmission across all areas in the short-term time scale of our model. ACT with long-acting partner drugs are currently in use in some areas; for example, artesunate-mefloquine is recommended by WHO. The more recently formulated dihydroartemisinin-piperaquine is also likely to become used more widely [47]. If used widely in the longer term in higher-transmission settings, the advantage of the prophylactic effects of these long-acting ACTs would need to be weighed against a higher selective pressure for resistant parasite strains [26,48,49]. Careful resistance surveillance would be important. Our finding that a long-acting nongametocytocidal drug regimen could have as much or more effect on transmission than a short-acting ACT in higher-transmission areas similarly needs to be considered bearing resistance development in mind. Such a drug regimen would need to contain two or more efficacious antimalarials in combination. Existing nonartemisinin combinations include SP-AQ and SP-chloroquine, but their use is limited due to widespread resistance [47]. However, the predicted impact of long-acting treatments has implications for future development of nonartemisinin combinations and their use as a first-line treatment, particularly if cost or development of artemisinin resistance make ACT a less attractive option. Developing and deploying drugs with stronger gametocytocidal activity than artemisinins would also be highly beneficial for transmission reductions according to our results (Figure VI in Text S1). Primaquine may be one such existing drug, but limited evidence does not show a consistent advantage over artemisinin derivatives [50,51].

Estimated treatment impact did not vary evenly with baseline slide-prevalence of infection, particularly in the medium-transmission settings (Figures 3A, A,4A,4A, and and5).5). This is likely to be due partly to sampling variation resulting from the smaller sample size in the area with 18.1% baseline prevalence (Table 1). The prevalence of symptomatic malaria and reported use of antimalarials are high in this area relative to the others given its measured infection prevalence, resulting in higher estimates of treatment impact. However, as has been found in other endemic populations [52], reported use of antimalarials also did not decline evenly with baseline prevalence in the other areas despite larger sample sizes. Variation in ACT impact due to treatment-seeking behaviour of populations is likely to be a real feature of ACT scale-up. Our results should be seen in terms of short-term impact only, since we do not take into account changes in the immunity of populations over time. The necessary data for our model, such as the incidence of infection and the proportion of infections developing symptoms, are rarely measured directly due to the difficulties of detecting all superinfections; therefore, we relied on model estimates from prevalence data, which gave results that were mostly within the range of the few data available [33,34]. Varying parameters and the model assumptions that affected these estimates caused limited change in the results except in the case of heterogeneity of exposure (see the sensitivity analysis in Text S1). In the highest-transmission setting the estimated incidence of infection and clinical episodes in the pre-ACT setting appear high (Table 3), but may be reasonable given measured entomological inoculation rates [23], and that the survey was conducted during the season of highest transmission and thus true annual rates would be lower. Whilst the prevalence of symptomatic malaria in the survey data decreased with transmission setting, our estimated proportion of infections developing symptoms did not show a clear trend of increasing as would be expected due to decreasing immunity (Table 3). This trend was visible across the four higher-transmission settings, but not in the two sites with lowest transmission intensity, probably due to the small numbers of cases meeting our definition of symptomatic malaria in these areas (Table 1 in the main text and Table II in Text S1). However, we had estimates of treatment intake in the population, and our results are relatively robust to whether this treatment occurs in symptomatic or asymptomatic individuals (Figure VIIa–b in Text S1). They are more sensitive to the accuracy of our treatment data, which is self-reported, and thus potentially underestimated [53]. Our data are limited to individuals under 45 y of age, which in these settings represents around 87% of the population [54]. By assuming in our model that individuals over 45 y have the same infection and immunity status as 15- to 45-y-olds, we may underestimate overall immunity levels in this age group to some extent. However, our sensitivity analysis showed that results were not affected substantially by ignoring age structure (Figure VIIi–j in Text S1). In generalizing our results to other settings it should be borne in mind that our slide-prevalence data represent the peak in annual transmission in these areas, whilst immunity levels in the study populations reflect lower transmission during other times of the year.

Our model predictions are roughly compatible with some of the trends from observational studies in areas where ACT was introduced, although it is uncertain to what extent these trends were attributable to ACT. In studies where transmission was similar to our lowest transmission setting (Kilimanjaro, >1,200 m) or lower, around 50% [10,12] to 75% [13] reductions in clinical episodes or prevalence occurred following ACT introduction, although there were also changes in vector levels during the same time period. In an area of Zanzibar with transmission levels similar to those of our second-lowest transmission setting (Kilimanjaro 600–1,200 m), little clear change in clinical incidence was seen during 2004 after the introduction of ACT, during which time 34,724 doses of artemether-lumefantrine were dispensed to a population of ~85,000 [14]. We cannot be sure what coverage of ACT this achieved (i.e. what proportion of all antimalarials used by the population were ACT), but in our equivalent setting this would amount to ~30% coverage and a potential ~16% reduction in clinical episodes (Figure 3B), which could have been masked by the seasonal trends in Zanzibar or diminished by imperfect patient adherence. As well as comparing our results to studies of ACT impact, it is interesting to compare with previous introduction of long-acting nonartemisinins, which received less attention in terms of transmission reductions. For example our model suggests that when SP was first widely introduced as an efficacious drug to replace failing chloroquine treatment, there could have been a 20%–30% reduction in clinical episodes in higher-transmission areas (Table 4) due to its long prophylactic time. Documented decreases of >40% in clinical episodes or hospital admissions have coincided with SP introduction in areas of intense transmission in southern Tanzania [55] and on the coast of Kenya [45,56], although as with ACT these studies were observational and took place in the context of increasing vector control. Cluster-randomized trials measuring the transmission impact of different antimalarials used as first line treatment would be valuable in order to confirm observational evidence and the findings of our model.